A High-Resolution Multifocal RGB Pollen Grain Image Dataset for Deep Learning Computer Vision Tasks from Biobío Region, Chile

PollenBB16 is an RGB pollen grain image dataset acquired by brightfield optical microscopy from botanically identified Chilean flora. Every annotation has been fully verified by an expert palynologist to guarantee taxonomic reliability at the instance level. The dataset comprises 16,198 images at the native sensor resolution of 3088 × 2064 px (8-bit per RGB channel, sRGB JPEG) and 36,383 pixel-accurate polygon instances distributed across 16 species in 13 botanical families. The species set spans three biogeographic groups relevant to the South American temperate forest: endemic (Eucryphia glutinosa, Quillaja saponaria, Otholobium glandulosum, Escallonia pulverulenta), native (Aristotelia chilensis, Gevuina avellana, Lomatia hirsuta, Schinus polygamus), and introduced species of apicultural and agronomic relevance (Anthemis cotula, Brassica rapa, Castanea sativa, Conium maculatum, Galega officinalis, Medicago sativa, Mentha pulegium, Rumex acetosella). Each spatial position on a slide is captured at three focal planes (fp0/fp1/fp2) along the optical axis, providing depth-varying diagnostic information (apertures, exine ornamentation, internal content) without the storage cost of a denser z-stack. Polygons are propagated identically across the three focal planes, since the external silhouette of the grain is invariant within the z-range.